Biochemistry 1994, 33, 4460-4470
4460
Articles
Structure of a
Zinc Finger Domain: Insights into ADR 1-UAS 1 Protein-DNA Recognition+,$ Histidine-X4-Histidine
Bradley E. Bernstein,i Ross C. Hoffman,rsll Suzanne Horvath,* Jon R. Herriott,s and Rachel E. Klevit'J Department of Biochemistry, SJ-70, University of Washington,Seattle, Washington 981 95, and Division of Biology, California Institute of Technology, Pasadena, California 91 I25 Received September 21, 1993; Revised Manuscript Received December 6, 1993"
ABSTRACT: The solution structure for a mutant zinc finger peptide based on the sequence of the C-terminal A D R l finger has been determined by two-dimensional N M R spectroscopy. The mutant peptide, called PAPA, has both proline residues from the wild-type sequence replaced with alanines. A nonessential cysteine was also replaced with alanine. The behavior of PAPA in solution implicates the prolines in the conformational heterogeneity reported earlier for the wild-type peptide [Xu, R. X., Horvath, S.J., & Klevit, R. E. (1991) Biochemistry 30, 3365-33711. The solution structure of PAPA reveals several interesting features of the zinc finger motif. The residue immediately following the second cysteine ligand adopts a positive 4 angle, which we propose is a common feature of this class of zinc fingers, regardless of whether this residue is a glycine. The N M R spectrum and resulting solution structure of PAPA suggest that a side-chain to side-chain hydrogen bond involving an arginine and an aspartic acid analogous to one observed in the Zif268 protein-DNA cocrystal structure exists in solution in the absence of D N A [Pavletich, N. P., & Pabo, C. 0. (1991) Science 252, 809-8171. A model for the interaction between the two ADRl zinc fingers and their D N A binding sites was built by superpositioning the refined solution structures of PAPA and A D R l b onto the Zif268 structure. This model offers structural explanations for a variety of mutations to the ADRl zinc finger domains that have been shown to affect DNA-binding affinity or specificity.
Zinc fingers constitute an important class of DNA-binding motifs in eukaryotes [see El-Baradi and Peiler (1991) for a review]. The motif was originally recognized as a repeating element in the sequence of the Xenopus transcription factor TFIIIA (Brown et al., 1985; Miller et al., 1985). The critical elements of the 30-residue structural domain include two cysteine and two histidine residues that serve as ligands for a Zn2+ion, giving the motifs their (Cysz, HisZ) nomenclature, and three conserved hydrophobic residues that interact in a small hydrophobic core. These stable domains consist of a Cys-Cys loop (two antiparallel strands connected by a reverse turn) and an a-helix containing the two histidine ligands, both free in solution, as defined by 2D-NMR' (Pgrraga et al., 1988;Lee et al., 1989b; Klevit et al., 1990;Omichinski et al., 1990; Kochoyan et al., 1991) and when bound to DNA in a t These studies were supported by NIH Grant PO1 GM-32681 . B.E.B. was supported by an NIH Medical Scientist Training Program (T32 GM-07266). R.C.H. was supported by an NIH Training Grant in Molecular Biophysics (T32 GM-08268). R.E.K. was supported by an American Heart Association Established Investigatorship. The top 10 structures and the average structure have been deposited in the Brookhaven Protein Data Bank under the filename IPAA. * To whom correspondence should be addressed. 8 University of Washington. 11 Current address: Department of Protein Chemistry, Immunex Research and Development Corp.,51 University St.,Seattle, WA 98101. 1 California Institute of Technology. e Abstract published in Advance ACS Abstracts, February 1, 1994. Abbreviations: 2D-NMR, two-dimensional nuclear magnetic resonance spectroscopy; NOE, nuclear Overhauser effect; NOESY, nuclear Overhausereffect spectroscopy;P.E.COSY,primitiveexclusivecorrelation spectroscopy; H-bonds, hydrogen bonds; Tris, tris(hydroxymethy1)aminomethane; COSY, correlation spectroscopy; RELAY, relayed coherence transfer spectroscopy; RMSD, root-mean-square deviation; TOCSY, total correlation spectroscopy.
0006-2960/94/0433-4460$04.50/0
cocrystal (Pavletich & Pabo, 1991). ADRl is a transcription factor that is required for expression of the gene for the glucose-repressiblealcohol dehydrogenase isozyme ADH2 in the yeast Saccharomyces cereuisiae (Denis &Young, 1983). TheminimalsequenceofADRl that retains high-affinity sequence-specific DNA-binding activity contains two (Cysz, Hisz) zinc fingers. Mutagenesis of this region has revealed that the sequence-specificinteraction with DNA is mediatedthroughthetwofingers (Thukraletal., 1991b, 1992). ADRl binds to a 22-base-pair palindrome upstream activator sequence (UAS1) in the promoter for the ADH2 gene. Two monomers of ADRl bind independently to the two identical half-sites within UASl (Thukral et al., 1991a). Both ADRl and UAS 1 have been subjected to mutagenesis experiments that have identified the amino acid residues and the base pairs that are critical to the protein-DNA interaction (Thukral et al., 1991b, 1992). Our studies have been aimed at determining the three-dimensional structures of the ADRl zinc fingers with the goal of providing a structural foundation on which to understand the functional properties of the protein. Our structural studies have focused on each of the two finger domains from ADRl as independent domains, defining their structures in solution using 2D-NMR spectroscopy. Although a single (Cys2, Hisz) zinc finger does not show detectable site-specific DNA binding, single fingers fold stably into a structure consistent with those reported for functional multifinger domains. Previous NMR analysis of the C-terminal finger from ADR1, ADRla, demonstrated that the peptide exists in multiple conformations (Xu et al., 1991). The conformationalheterogeneitymanifests itself as peak doubling in spectra of ADRla. This peak doubling has made a quantitative determination of the wild-type ADRla solution 0 1994 American Chemical Society
Biochemistry, Vol. 33, No. 15, 1994 4461
Structure of a His-X4-His Zinc Finger Domain structure difficult, and we have therefore confined ourselves to a description of the overall folding topology of this domain. In contrast, the N-terminal ADRl finger peptide, ADRlb, yields a single set of lines in its NMR spectrum, and its threedimensional structure has been determined and refined (Klevit et al., 1990; Hoffman et al., 1993a). Although both ADRl zinc fingers conform to the minimal zinc finger sequence (Michael et al., 1992), there are two notable differences in their sequences (refer to Figure 5 for sequences). First, ADRla contains prolinesat positions 131and 133 (numbering used is for the ADRl protein), whereas ADRlb contains no prolines. Second, ADRlb is a His-X3-His finger (Le., it has a three-residue spacing between the zinc-coordinating histidines), while ADRla is a His-X4-His finger. It was hypothesized that one of the aforementioned properties is responsible for the conformational heterogeneity observed in the ADRla peptide (Xu et al., 1991). In order to test the contribution of the proline residues to the observed ADRl a peak doubling, and in hopes of obtaining clear structural data for the ADRla domain, a mutant in which both prolines were substituted with alanine (called PAPA) was synthesized. This peptide also contained a Cys140-to-alanine substitution. Since this cysteine is not a zinc ligand and has been shown by alanine-scanning mutagenesis to be noncritical for DNA binding (Thukral et d., 1991),alanine was substitutedat position 140to yielda peptide that is less susceptible to oxidation. The results of twodimensional ‘H NMR studies of the PAPA peptide are reported here. The refined solution structure of this domain is described and related to the known functional properties of this finger domain in ADR1. MATERIALS AND METHODS Sample Preparation. A peptide corresponding to the C-terminal zinc finger of ADR1, but with the modifications listed below, was synthesized and purified as previously described (Pirragaet al., 1990). The peptidesequenceconsists of ADRl residues 130-159 with alanine substitutions at residuescorresponding toPro131,Pro133, and Cysl40in wildtype ADR1. The sequence of the resulting mutant, PAPA, numbered as in ADRl protein, with the three non-wild-type residues underlined is 130
159
NH3+-K AYACGLCNRAFTRRDLLIRHAQKIHSGNK -COO-
The peptide was analyzed for purity and correctness by analytical HPLC, mass spectrometry,and Edman degradation sequence analysis. NMR samples were prepared in 50 mM deuterated Tris and 25 mM deuterated acetate, pH 5.29 (meter reading, uncorrected for isotope effect), as described previously (Hoffman et al., 1993a). NMR Methods. NMR experiments were performed on a Bruker AM-500 spectrometer. The sample temperature was held at 25 OC, and the spectral width was 10000 Hz. Presaturation of the water resonance during the 2.0-s relaxation delay suppressed the solvent signal. 2D spectra were collected as previously described (Hoffman et al., 1993a,b). Felix I1 (Hare Research, Woodinville, WA) was used for data processing. Proton resonances were assigned from COSY, TOCSY, and NOESY experiments by standard practice (Wiithrich, 1986). P.E.COSY spectra collected on a D2O sample were used to determine 3Jap coupling constants (Mueller, 1987). These were used in conjunction with NOESY intensities to obtain stereospecific assignments for prochiral Cp methylene
protons and leucine C6 methyl protons where possible as described previously (Hoffman et al., 1993a). Relative chemical shifts (A) used to compare PAPA to other zinc fingers were calculated by subtracting the randomcoil chemical shift for the relevant amino acid type (Wiithrich, 1986) from the observed (or reported) chemical shift. Comparisons were performed using residues from the first conserved aromatic residue to the first conserved histidine residue in each finger sequence. Structure Calculation and Refinement. Two identical NOESY sets collected with 100-ms mixing times were used to generate distance constraints. The spectra were processed as described previously (Hoffman et al., 1993b). The spectra were added together, and the resulting intensities were measured using volume integrals from the additive spectrum. A diagonal-suppressed NOESY spectrum was used to identify cross peaks that were close to the diagonal (Denket al., 1985). Dihedral angle constraints were included in the structure calculation when unambiguous assignments could be made. Sixteen x1 rotamer assignments were made on the basis of coupling constants and/or relative intraresidue NOESY intensities. In addition, x2 constraints for Leul 46 and Leul 47 could be made on the basis of the stereospecific assignments of the leucine Cpprotons and the relative intraresidue NOESY cross peaks. No H-bond constraints or tetrahedral zinc constraints were used in the structure calculations. Lists of the constraints used as input for the structure determination were made available to the reviewers and are available as supplementary material. Solution structures for PAPA were determined using the refinement method called REPENT (Hoffman et al., 1993b), in combination with DIANA (version 1.O) distance geometry software (obtained from K. Wiithrich, ETH, Zurich) and the complete relaxation matrix algorithm CORMA (obtained from T. James, UCSD, San Francisco), as previously described (Hoffman et al., 1993a). In this approach, iterative rounds of refinement are performed to obtain structural models that are more consistent with the experimentally measured NOE intensities, as measured by R(1/6), where R(1/6) = C{((lobs)1/6 - (~calc)1’61)/c(lobs)1/6. Structural Analysis. Statistics reported in Table 3 were calculated as described previously (Hoffman et al., 1993a). RMSDs were calculated over residues 132-1 55 and hydrogen bonds predicted for the 10 best structures generated (i.e., the structures with the lowest R(1/6)). An average of the 10 structures was calculated and used for subsequent analysis. The top 10 structures and the average structure have been deposited at the Brookhaven Protein Data Bank. Structures were visualized and compared using QUANTA (Polygen, Corp., Watham, MA) and INSIGHT I1 (Biosym Technologies Inc., San Diego) on a Silicon Graphics workstation. Individual fingers were superpositioned and RMSDs between them calculated using the backbone atoms for residues spanning from the first conserved hydrophobic (analogous to Tyr 132 in PAPA) to the first coordinating histidine residue. This method afforded the closest correspondence between HisXs-His and His-XpHis fingers. In addition, models of the presumed interaction between the ADRl zinc fingers and their DNA subsites were created. These models, composed of a single ADRl finger docked to its respective DNA-binding subsite, were created by superpositioning either the PAPA or the ADRl b average structure on finger 3 of the Zif268 proteinDNA coordinates and by replacing the relevant DNA bases in the Zif268 binding site with their UAS 1counterparts. These superpositions were achieved using the backbone atoms only,
4462 Biochemistry, Vol. 33, No. 15, 1994
Bernstein et al.
I
10.0
8.0
4.0
6.0
2.0
0. c
PPm
FIGURE1: IH NMR spectrum of PAPA. One-dimensional spectrum of PAPA peptide dissolved in 50 mM deuterated Tris and 25 mM deuterated acetate, pH 5.29 (meter reading, uncorrected for isotope effect), in DzO.The peptide was reconstituted in the presence of 1 . 1 mol equiv of ZnZ+.
starting with the first aromatic residue and ending with the first coordinating histidine residue.
RESULTS AND DISCUSSION Single-Finger Peptides Assume Multiple Conformations. Previous studies of ADRla demonstrated that the peptide exists in more than oneconformation that interconverts slowly on the NMR time scale, as evidenced by an observed peak doubling in its NMR spectra (Xu et al., 1991). Analysis of the spectra indicated the presence of two folded conformations of ADRla, a major form present in -70% and a minor form present in -30% of the population. Since the presence of proline residues has been shown to be responsible for slowly interconverting species in other proteins, the mutant peptide PAPA was designed in which both proline residues of ADRla were replaced by alanine. Unlike the spectrum of ADRla, the 1D spectrum of PAPA (Figure 1) does not show evidence of peak doubling, as is clear from the single sharp resonances of the two histidine C' protons at -8.0 ppm. However, when 2D spectra of PAPA with extremely high signal to noise were analyzed, a small amount of peak doubling was detected. Similar handling of spectra obtained for the other finger sequence from ADR1, ADRlb, did not reveal peak doubling. We must therefore conclude that this property is peculiar to the ADRla sequence, rather than due to some sort of artifact of peptide synthesis or purification.2 The PAPA spectrum indicates that removal of the two prolines from the ADRla sequence has a significant effect on the equilibrium between the two forms but does not completely abolish either conformation. Clearly, further mutants of ADRla must be designed in order to understand this phenomenon more fully. However, the spectral properties of PAPA are good enough to allow a detailed structural analysis of this mutant peptide. As a control, the single mutant ADRla:C140A was synthesized and its NMR spectrum analyzed. Like wild-type ADRla, the spectrum of this peptide displays significant peak multiplicity (D. English and R. Klevit, unpublished results). This rules out the nonligand cysteine at position 140 as the source of spectral heterogeneity originally reported for ADRla.
The Zinc Finger Signature Fold. Complete assignments were made for PAPA using conventional 'H homonuclear N M R techniques and are listed in Table 1 (see Materials and Methods). A strong correlation has been noted between N H and CaH resonance frequencies of analogous residues of (Cys2, His2) zinc finger peptides (Michael et al., 1992; Lee et al., 1992). Such a correlation suggests that it is mainly the backbone conformation at each residue, rather than theidentity of the side chain, that defines its resonance frequency with respect to random-coil values. Thus, the backbone proton resonance frequency shifts can serve as a signature for the zinc finger fold. With this in mind, the N H and C"H shift values of PAPA were compared residue by residue with those of several other (Cysz, Hisz) zinc finger peptides. As shown in Figure 2, there is remarkable correspondence between PAPA and the seven fingers to which it was compared. The correlation coefficients for pairwise comparisons are listed in Table 2. Not surprisingly, the strongest correlations are between PAPA and ADRla, its parent sequence. There is also significant correlation when the resonance frequencies of PAPA are compared with the other ADRl finger peptide, ADRlb (Hoffman et al., 1993a) and with the peptide based on one of the finger domains from the Xenopus protein, Xfin (Lee et al., 1989a). Both ADRl b and Xfin are classic "odd" fingers as dictated by their His-X3-His spacing and the position of the second conserved aromatic (Kochoyan et al., 1991). In contrast, ZFY-6 is an "even" finger with a His-X4-His spacing and its conserved aromatic two residues N-terminal from the odd position. PAPA is a hybrid, with the aromatic position of an "odd" finger and the His-Xd-His spacing of an "even" finger. On the basis of its NH shift values, PAPA corresponds more closelyto the odd fingers analyzed. Moreover, although the additional residue between the histidines in PAPA causes the N H fingerprint between the coordinating histidines to diverge, the histidine resonances themselves conform closely to those in conventional odd fingers. Interestingly, PAPA corresponds more closelyto the minor form of ADRla on the basis of resonance shifts. The largest chemical shift difference between the two forms of wild-type ADRla is observed for the amide proton of Tyr132 (-0.8 ppm), the residue between the two prolines (Xu et al., 1991). The chemical shift for the N H of Tyr 132in PAPA corresponds almost exactly to its value in the minor form of ADRla. As illustrated in Figure 2, the amide proton of the first aromatic residue is shifted significantly upfield relative to the randomcoil position in most of the zinc fingers that have been analyzed. The minor form of ADRla and PAPA also show this large upfield shift, leaving only the major form of ADRla and MBP1bas exceptions. Thus, if the backbone frequencies are indeed signatures for the backbone conformation of the zinc fingers, these correlations suggest that PAPA mimics the minor form structure of ADRla in solution and therefore that replacement of one or both of the prolines causes the equilibrium to favor the so-called minor form. The ADR 1a finger resonances in the context of a two-finger synthetic peptide also correspond moreclosely to the minor form, suggesting that PAPA contains structural elements of the multifinger DNA-binding ADRlc construct (R. X. Xu and R. E. Klevit,unpublished observation). Analysis of the well-resolved and well-defined NMR spectra of PAPA helped to clarify difficult assignments in ADRla wild-type and T142I peptides (Xu et al., 1991). Specifically, theC"HassignmentsforCysl40 (A =-0.10ppmwithrespect to random coil) and Arg144 (A = 0.10 ppm) in the original report are not consistent with the analogous assignments in
Biochemistry, Vol. 33, No. 15, I994
Structure of a His-Xd-His Zinc Finger Domain
4463
Table 1: Proton Chemical Shifts (in ppm)O for PAPA at 25 OC and pH 5.5 residue
NH
C*H
CBH
CrH
C6H
CfH
CrH
1.75 1.21 1.60 3.84 Lys 130 1.18 4.44 8.31 Ala131 8.74 4.52 6.99 6.81 Tyr132 2.74: 2.92 4.90 1.29 8.64 Ala133 2.79: 3.33 8.88 4.40 cys134 4.16,3.90 9.28 Gly135 0.60: 1.04 1.37 0.68,c 0.50 4.41 8.95 Leu136 3.14: 3.31 7.99 4.97 cys137 7.51, 6.79d 2.95 4.68 8.11 Asn138 1.38: 1.24 1.68, 1.45 3.10,3.02 7.14c 4.11 Arg139 8.01 1.15 4.89 Ala140 7.95 2.80: 3.28 6.16 7.24 6.82 8.48 4.91 Phel41 1.33 4.49 4.55 Thr142 9.02 1.72 1.57:2.13 3.28 8.43C 4.7 1 Arg 143 7.06 1.37, 1.27 3.08 7.2V 1.51, 1.12 3.09 Arg144 8.54 2.55: 2.65 4.10 Asp145 8.67 1.61 0.86; 1.11 1.93: 1.49 Leu 146 6.88 3.98 1.58 1.01; 0.99 1.30: 2.06 3.28 Leu147 6.92 1.79 1.51, 1.22, 0.83 (m) 0.73 8.05 3.61 Ile148 1.79 7.25c 1.65 3.16 7.67 4.00 Arg 149 8.01 7.10 2.83: 3.26 7.52 4.27 His 150 1.59 8.79 3.81 Ala151 2.19: 2.02 2.55,2.36 7.38, 6.74e 8.11 3.81 Gln152 1.46, 1.37 1.68 2.95 1.88, 1.79 7.85 4.1 1 Lys153 0.66 1.29,0.28 (m) 0.93 4.08 Ilel54 8.23 6.54 7.93 3.24: 3.37 7.25 5.23 His155 4.02, 3.95 4.27 Ser 156 8.06 8.47 4.05, 3.58 Gly157 2.70: 2.80 7.48, 6.95d 8.01 4.66 Ad58 1.59 1.22 0.87 7.71 4.25 Lvs159 0 All chemical shifts are reported relative to sodium 3-(trimethylsilyl)propionate-2,2,3,3-d~ (TSP). Stereospecific assignment, first chemical shift assigned to HB2. 0 Stereospecific assignment, first chemical shift assigned to CDl. N6H resonances. e NeH resonances.
PAPA. These residues were also noted as being inconsistent with chemical shifts reported for other zinc fingers (Lee et al., 1992). However, the analogousvalues in PAPA (A140, A = 0.54 ppm; R144, A = -1.29 ppm) are consistent with these reports. Since we suspectedthat the cysteineresidue at position 140 might be in part responsible for the poorer quality of the spectra of ADRla and T142A, a new “wild-type” ADRla peptide was synthesized in which Cysl40 was replaced by alanine. Although the spectra of this peptide still exhibit strong peak doublings, the quality of the spectra is indeed superior, making assignments somewhat clearer (D. English and R. Klevit, unpublished data). The assignmentsfor residues 140 and 144 in C140A (A140, A = 0.36 ppm; R144, A = -1.28 ppm) are consistent with those in PAPA and therefore with other zinc finger peptides. Even with these assignments in hand, it is difficult to find the corresponding resonances in the original ADRla spectrum. Since the existenceof a cysteine at residue 140 in ADRla and T142I may perturb the C‘H shift relative to the other peptides, it remains unclear whether this assignment is correct as originally reported. On the basis of this analysis, however, we conclude that the original assignment for the CuH of residue 144 in ADRla was most likely incorrect. PAPA Structure. The solution structure of the PAPA peptide was calculated from two-dimensional lH NMR data, as described in Material and Methods. A region of an HzO NOESY spectrum used to generate distance constraints was made available to reviewers and is available as supplementary material. The 10 structures with the lowest residual error function R(1/6) for observed vs calculated NOE intensities were used for structural analysis (see Materials and Methods); their statistical parameters are presented in Table 3 and superpositions are shown in Figure 3. The backbone structure of the family is very well defined by the NMR data, as illustrated in Figure 3A and indicated by the average RMSD
of 0.21 A (calculated over residues 132-155). Many of the side-chain conformations are also well-defined in the PAPA solution structure (Figure 3B,C). Only 6 of the 30 side chains have average RMSDs greater than 1.O A and all of these are hydrophilic, surface residues: Asn138, Arg139, Arg143, Arg144, Arg149, and Lys153. The solution structure of PAPA reiterates all the structural elements previously described for (Cysz, His$ zinc fingers: a Cys-Cys loop consisting of two antiparallel strands connected by a reverse turn, the “fingertip” in which the backbone makes a 90’ turn, and a helical region that includes the histidine ligands. The PAPA backbone and zinc coordination geometry closely resemble the refined models for ADRlb (Figure 4), as well as for other experimentally determined zinc finger domains (Hoffmanetal., 1993a; Leeetal., 1989b;Omichinski et al., 1990; Kochoyan et al., 1991). The backbone RMSD between the average structures for ADRlb and PAPA is 0.46 A (calculated as described in Materials and Methods). The remarkable similarity of PAPA to ADRlb (as well as to Zif268, see below) confirms that the sequence of ADRla is capable of taking on a canonical zinc finger structure when there are alanines at positions 131 and 133. Although the structure of PAPA per se cannot directly address the role of the naturally occurring prolines, it can address other important aspects of the ADRla structure, in particular, the effect of the alternate His-Xd-Hisspacing. Since a detailed description of the refined solution structure of ADRlb as calculated by an identical protocol has been published recently (Hoffman et al., 1993a), we will focus mainly on a comparison of the solution structures of PAPA and ADRlb. In addition, structural characteristics will be compared to those in the X-ray structure of the three-finger Zif268 protein (Pavletich & Pabo, 1991). Cys-Cys Loop. The N-terminal residues of PAPA form the Cys-Cys loop, consisting of two antiparallel strands and
Bernstein et al.
4464 Biochemistry, Vol. 33, No. 15, 1994
A
1-
@A1
mn X
0
A
0-
0
0
-1
A
;
h 0
@
$1
0 A
a A
A
0
ADRlb
x
Xfin
0 ZFY-6 (Even) A MBP-la A MBP-lb
X
0
:
I
I
I
-2
-1
0
1
-2
ADRla (minor)
0 ADRla (major)
H 3A
A
8
-
A
X
0
0
O A
d3
W
% A
A
PAPA NH Relative Shifts (ppm)
ADRla (minor) ADRla (major) PAPA
0
t 3
ADRlb Xfin
lam MBP-la III] MBP-lb
-0.6
i
I
First Conserved Aromatic
Residue
FIGURE2: Relative chemical shifts for N H resonances of a number of zinc finger sequences. The relative chemical shifts were calculated as described in Materials and Methods. (A, top) The correspondencebetween the shifts for PAPA and seven other zinc finger sequences are plotted for all residues between and including the first conserved aromatic residue and first conserved histidine residue. (B, bottom) The relative chemical shifts for the NH of the first conserved aromatic residue are shown. For references regarding the chemical shifts, see legend to Table 2. a reverse turn. The two positions where proline side chains are found in the wild-type sequence have backbone dihedral angles that are consistent with a @-strandconformation: the average (4, J / )values for Ala131 and Ala133 are (-90 f 50°, 140 f 15') and (137 f 6', 119 f 12'), respectively. (The wide range of 4 values in the solution structure for Ala 131 is likely due to its proximity to the N-terminus of the peptide.) Proline is limited to 4 values of -60 f 20' and prefers J/ valuesof --So or 145'. Thus,ofthetwomutatedresidues in PAPA, the solution structure clearly indicates that Ala133 exists in a conformation that is not accessibleto the wild-type proline. Two of the fingers in Zif268 contain a proline at the position analogous to Pro131 in ADRla, both of which are in an allowed (4, J / )conformation of ( --60°, --loo). None of the Zif268 fingers contain proline at the position analogous to Pro133. In all three fingers, the residue in this position
-
takes on (6,J / ) values of (--1 loo, 145'), consistent with a @-strandconformation and with the value observed for Ala 133 in PAPA. Taken together, these results indicate that, in the absence of proline residues,this region of the zinc finger domain exists in an extended @-strandconformation, which may be deformed slightly to accommodate a proline in the position analogous to Pro13 1 in ADR 1a. The consequence of having a proline at position 133 must await a thorough structural analysis of a finger with such a proline residue in place. However, it appears that adjustments of the backbone will be required to accommodate a proline at this position. Similar to ADR 1b, cross-strand H-bonds are predicted between the carbonyl of Tyr 132 and the amide of Phel41 and NH-sulfur H-bonds analogous to those reported for ADRl b are predicted (NH136 and NH137 to the Sr of Cys134 and N H 139 to Cysl37). Also analogous to the solution structure
Structure of a His-X4-His Zinc Finger Domain Table 2: Chemical Shift Correlations between Various Zinc Finger Domains'
Biochemistry, Vol. 33, No. IS, 1994 4465
4
ADRla ADRla majorb minorb ADRlbc MBP-lbd Xfin31) ZFY-6/ 0.91 0.96 0.84 0.68 0.65 0.41
PAPA ADRla major 0.96 0.78 0.74 0.62 0.47 ADRla minor 0.82 0.67 0.67 0.49 a Relative chemical shifts were calculated as described in Materials and Methods for the N H resonances for residues from the first conserved aromatic through the first conserved histidine residue. Pairwise comparisons between zinc fingers were performed by plotting the relative chemical shifts for each corresponding residue and fitting the resulting plot to a linear function. The numbers in this table are the correlation coefficients for the pairs of data [see, for example, Kurtz (1963)l. b Chemical shifts used are for the ADRla peptide in which a nonessential cysteine was replaced with alanine, C140A (D. English and R. Klevit, unpublishedresults). AsreportedinHoffmanetal. (1993). Asreported in Omichinski et al. (1992). As reported in Lee et al. (1989a). /As reported in Kochoyan et al. (1991). Table 3: Structural Parameters for PAPA parameter av' DIANA target function! A2 RMSDc (N, Ca, C, 0),A RMSDC (all heavy atoms), A av'R(1/6), % no. of interproton distance constraintsd intraresidue i,i+ 1 i,i+2 i,i+3 i,i+4 long range total no. of dihedral angle constraintsC
PAPA 4.1 f 0.4 0.21 f 0.06 1.13 f 0.18 9.9 f 0.1 159 83 15 19 10 50 346
XI
16
x2
2
a Averages are reported for the 10 refined structures in the family. The target function is a residual error function based on upper and lower bound, dihedral angle constraint, and van der Waals violations for each calculated structure. The target functions reported here were calculated with weighting factors set at 1.0 for each of upper distance bounds, lower distance bounds, and van der Waals and set at 5.0 for dihedral angle constraints. All RMSDs are reported as the average pairwise RMSDs among the 10 refined structures for the residue range 132-155. dThe upper and lower bound files used in the structure calculations are available as supplementary material. e The dihedral angle constraints used in the structure calculations are available as supplementary material.
of ADRl b, the residue following the second zinc-coordinating cysteine residue (Asnl38 in PAPA; Thr 110 in ADR 1b) has a positive #J angle. The 4 and # angles adopted by Asn138 (89O and -14O, respectively) are consistent with a G1 fl-bulge in which a similarly positioned glycine has (4, #) values of ( 8 5 O , Oo) (Richardson, 1981). As in ADRlb, additional H-bonds predicted between the amide of Cysl34 and the carbonyl of Arg139 and between the carbonyl of Cys134 and the amide of Asn138 may help to stabilize this geometry. Although the positive #J angle has not been explicitlymentioned in reports of other zinc finger structures, we wondered whether it is a common feature. In (Cys2, His2) zinc fingers, the residue that follows the second cysteine ligand is most often glycine (Michael et al., 1992). In addition to the positive #J angles in the two ADRl fingers, an analogous non-glycine residue in Zif268 finger 2 (Metl41) also has a positive #J (Pavletich & Pabo, 1991). Moreover, we note that, in all rubredoxin structures solved to high resolution by X-ray diffraction, the residue that follows the second cysteine ligand in its Cys-Cys loop (which is always glycine) assumes a positive $I angle (Adman et al., 1991.) It is therefore likely that the positive
FIGURE 3: Solution structure of PAPA. (A, top) Superposition of the top 10 structures, showing backbone atoms only. All residues of PAPA (Le., 130-159) are included. (B, middle) Superposition of the top 10 structures, with all heavy atoms included. For clarity, only residues 132-155 are shown in this view. (C, bottom) Superposition of the top 10 structures, showing side-chain heavy atoms for the seven conserved zinc finger residues (four zinc ligands and three hydrophobics). In addition, the side chains of Arg143 and Asp145 are shown.
Bernstein et al.
4466 Biochemistry, Vol. 33, No. 15, 1994 1
i
lAla E ElA A R 1i o
T
El V
\ Zn
T
a: +3
L
140
g
L +3
L
R
I
N
R +6
El
\
Y
/ H A Q
G O / z n \ H
V
s F
y
El c
K f l P
E
103
'
K
S G
130
ADRl b
ADRla
ADRla
ADRlb
nn +6 +3 -1 +6 +3 -1
R L
R R H R + + + + I O
S T T G G A G Y
FIGURE5: Summary of the results of mutagenesis experiments on ADR1. (A, top) The residues where substitution with an alanine resulted in a >lO-fold decrease in DNA-binding affinity are boxed in the sequence of the two zinc fingers of ADRl (Thukral et al., 1991b). (B, bottom) Summary of the protein-DNA contacts revealed from change-of-specificity mutagenesis (Thukral et af., 1992).
-1
h CTV
+3
His126/15F;r
'P
FIGURE 4: Comparison of the solution structures of PAPA (blue) and ADRl b (red). The average structures calculated from the solution structures for each peptide are shown, superimposed as described in Materials and Methods. (A, top) Side-chain atoms for the seven conserved zinc finger residues are included. Note that the main difference between the two structures occurs in the backbone between the two histidine ligands (lower right region of this view). (B, middle) The a-helices of ADRl b and PAPA superimposed. The side chains in the three putative DNA-binding positions (-1, +3, and +6) plus the second histidine residue are shown. Note that the long, hydrophilic side chains at positions -1 and +6 are not well-defined in solution. (C, bottom) Same as (B) but viewed down the helical axis. This view makes clear the tight loop between the histidines in A D R l b and the loose loop in PAPA.
4 angle assumed by the residue immediately following the second Cys is a feature common to many if not all of the Cys-Xz-Cys zinc finger structures. This arrangement brings the two cysteine side chains into proximity to bind the zinc ion, and the stabilization afforded by zinc binding likely serves to offset the unfavorable dihedral angle at this position. a-Helix. The a-helical regions of PAPA and ADRl b are shown in Figure 4. Alanine mutagenesis studies revealed that residues at analogous positions in the helical regions of the two ADRl fingers do not exhibit analogous functional properties (Thukral et al., 199lb, and Figure 5 ) . It is therefore of particular interest that the PAPA structure differs most markedly from ADRl b in this region. A superpositionof the PAPA and ADRl b structures, together with dihedral angle information, demonstrates that the helix in each of the molecules begins at an analogous residue (Arg144 in PAPA and Glnll6 in ADRl b) and ends at the second histidine. The backbones diverge only between the two histidine ligands: in order to approximate the coordinating histidines to the zinc ion, the three interceding residues in ADRl b tighten to form a 310 helix, while the four residues between the histidines in PAPA loosen such that the dihedrals for Gln152 and Lys153 are in altogether nonhelical geometries. These results appear to be consistentwith those observed in NMR studies of another His-X4-His zinc finger peptide, ZFY-6 (Kochoyan et al., 1991). A Hydrogen Bond between the Side Chains of Argl43 and Asp145. The Ne proton of Arg143 is shifted -1.2 ppm downfield (to 8.43 ppm) from the other arginine Neprotons in the PAPA spectrum, which resonate close to the randomcoil position of 7.2 ppm (see Table 1). A similar downfield shift is observed in the wild-type ADRla spectrum and in the T142I mutant spectrum (Xu et al., 1991). A downfield shift of a nitrogen-bound proton can be indicative of involvement in a H-bond. Interestingly, in the crystal structure of Zif268 the Ne protons of the analogous arginines in fingers 1 and 3 are each involved in a H-bond to a carboxylate oxygen of the aspartic acid residue that is two residues to its C-terminal side, forming a H-bond network involving the two side chains and a guanine in the cocrystal (Pavletich & Pabo, 1991). The
Structure of a His-XrHis Zinc Finger Domain analogous residue in PAPA is Asp145, which exists in a defined side-chain conformation in solution, as evidenced by its inequivalent CPH resonances, its 3Jap couplings, and its NOESY pattern. Thus, the spectral properties of Arg143 and Asp145 suggest that a similar H-bond interaction occurs in PAPA and in ADRla, even in the absence of DNA. The PAPA solution structure is consistent with this conclusion in that an Asp145 carboxylate oxygen and the Ne proton of Argl43 are in proximity to make such an interaction in a majority of the structures (see Figure 3C). It is important to note that the PAPA solution structures were calculated using only distance, dihedral angle, and van der Waals constraints. Thus, the proximity observed is a consequence of the experimental constraints placed on these two side chains, rather than on an electrostatic potential enforced during the structure calculation. The analogous residues in ADRlb are Argll5 and Glu117, which may be capable of making a comparable interaction. However, the arginine Ne proton shift was not observed for this peptide, and the glutamic acid side chain does not take on a preferred orientation in solution (Hoffman et al., 1993a). Either the longer glutamic acid side chain is sterically unable to make an analogous H-bond or the additional loss of entropy that would accompany its formation must overridethe small gain in enthalpy from the ionic H-bond. Structure-Function Relationships: Modeling of A D R l DNA Interactions. An ultimate goal of structural comparisons between ADRl fingers is to yield insight into the function of the molecule. The residues in the two zinc fingers of ADRl for which substitution led to a significant diminution of DNAbinding affinity are illustrated in Figure 5. Consistent with the hypothesis that both zinc liganding and the interaction of conserved hydrophobic residues are responsible for stable tertiary structure in a zinc finger motif, substitution of the conserved residues resulted in decreased DNA binding. The three-dimensional structures of the ADRl fingers, and all other fingers of this class that have been reported, confirm the contribution of the seven conserved residues to the tertiary structure. Therefore, we will focus on the mutations in nonconserved positions that affect DNA binding. When we conducted these studies, there was only one experimentally determined structure for a zinc finger-DNA complex, that of the Zif268 three-finger fragment bound to an 11-base oligonucleotide (Pavletich & Pabo, 1991). The Zif268 structure provides a detailed model of how zinc fingers can interact sequence-specifically with DNA. For this structure to serve as a guide for understanding ADR1, two criteria should be met: (1) the structures for the individual ADRl finger domains should be similar to the finger structures in the Zif268 cocrystal, and (2) there should be evidence to suggest that the mode of DNA recognition used by ADRl is similar to that used by Zif268. To ascertain whether the first criterion is met, main-chain RMSDs for the 19 residues spanning the first conserved aromatic to the first coordinating histidine were calculated between each ADRl finger and each Zif268 finger, as listed in Table 4. There is remarkable similarity between the ADRl and Zif268 fingers, even though the structures weredetermined by different techniques, using individual domains with no DNA present in the former case and multiple domains bound to DNA in the latter. Therefore, this criterion is satisfied. In the Zif268-DNA structure, three positions in the zinc finger make direct contact with DNA bases. For consistency, these positions are called -1, +3, and +6 with respect to their position relative to the a-helix. Each of the three fingers in Zif268 uses two of these three possible contacts. As illustrated
Biochemistry, Vol. 33, NO. 15, 1994
4467
Table 4: Main-Chain RMSDs between ADRl and Zif268 Zinc Finger Domains’ Zif268 f2 Zif268 f3 ADRlb PAPA Zif268 f2 0.418 0.613 0.655 Zif268 f3 0.657 0.644 ADRlb 0.462 PAPA RMSDs were calculatedfor the backbone atoms of residuesfrom the first conserved aromatic through the first conserved histidine residue, as described in Materials and Methods.
in the schematic in Figure 5 , alanine mutagenesis of ADRl indicates that all three positions are critical for DNA binding in ADRlb while only the -1 position is critical in ADRla (Thukral et al., 1991b). DNA mutagenesis studies indicated that each ADRl finger contacts a 3-bp “subsite” within the 11-bp UASl half-site: ADRlb contacts bases 8-10 (GAG) and ADRla contacts bases 5-7 (TTG). This arrangement, including the polarity of the complex, is also consistent with that observed in the Zif268-DNA complex. Finally, changeof-specificity mutagenesis of ADRl identified specific sidechain-base interactions, as shown schematically in Figure 5 . The above analyses suggest that ADRl follows the general theme described by the Zif268 finger-DNA interactions and therefore the second criterion is satisfied. As just demonstrated, the zinc fingers of ADRl appear to behave generally similar to those in Zif268. However, the mode of DNA binding for ADRl appears to be different in detail. In particular, the +6 DNA-contact position in ADRla does not mimic analogous residues in either the Zif268 fingers or in ADRl b. The +6 positions in ADRl b and Zif268 fingers 1 and 3 are all arginines that each contact guanine. Although ADRla also contains an arginine at the +6 position, the relevant base in its UASl binding site is thymidine (Thukral etal., 1992). Whilereplacement ofthe +6arginineinADRlb (Argl21) resulted in > 100-fold decrease in DNA binding affinity of a two-finger ADRl construct, the analogous replacement in ADRl a had no detectable effect (Thukral et al., 1991b). A comparison of the helical regions and the two +6 arginines of ADRlb and PAPA is shown in Figure 4. Although the positions of the ends of these surface-exposed residues are not defined in solution, the superposition shows clearly that the Ca positions of Argl21 and Arg149 are in identical positions relative to the rest of the helix so that the side chains extend from identical positions on the helices of ADRl b and PAPA. This suggests that the cause of the difference in the functional properties of Argl2l and Arg149 is not due to structural differences at the site in question and must therefore be due to a more distal effect. In hopes of gaining insight into the functional differences between the two ADRl fingers, a model for ADRl zinc fingerDNA interactions was generated. The average refined structure of each ADRl finger was simply superpositioned onto Zif268 finger 3 to give the lowest main-chain RMSD, and each 3-bp subsite was “mutated” into the relevant sequence from UASl without altering the DNA conformation (see Materials and Methods). As is clear from Figure 3B, the exact positions and conformations of the side chains of surface residues (especially long hydrophilic residues such as arginines and lysines) are not well-defined in solution and are therefore not likely to be representative of their position or conformation when bound to DNA. However, the backbone positions of these residues are defined in the NMR structures and can be considered indicative of their placement in the DNA-bound
4468 Biochemistry, Vol. 33, No. 15, 1994
Bernstein et al.
FIGURE6: Models for the interaction between ADRl and UAS1. Models were generated by superpositioningthe solution structures of PAPA (blue) and ADRlb (red) on the structure of Zif268, as described in Materials and Methods. The putative contacts for position +3 side chains are shown for each finger. (Left) Superposition model for the ADRla finger. Note the steric clash between the methyl group of Leu146 and the thymidine methyl. (Right) Superposition model for the ADRlb finger. His1 18 is in excellent distance and geometry to participate in an H-bond with the adenine. protein. The ADR1-DNA models place the three residues at the -1, +3, and +6 positions of the a-helix facing into the major groove of the DNA, poised to make the interactions predicted from the ADRl mutational analyses. In addition to positioning the putative base contacting side chains reasonably with respect to the bases themselves, the ADR1-DNA models also retain the intriguingcontact between the N6 of the first zinc-coordinating histidine to a backbone phosphate observed in all three Zif268 fingers. This contact was hypothesized to serve to properly orient finger domains with respect to their DNA sites (Pavletich & Pabo, 1991). While it is not a surprising result for ADRlb, the PAPADNA model confirms that such an interaction can also anchor a His-X4-His finger to its binding site in an orientation similar to the more canonical His-X3-His fingers. In addition, although the histidine ligand is in proximity to the DNA backbone, the PAPA-UAS 1 superposition model does not indicate any clashes between the large His-X4-His loop and the DNA. The fact that both ADRl finger-DNA models affirm central themes predicted from other structural and functional data led us to look for novel findings. We therefore used these models to try to understand why, in contrast to the N-terminal finger, the +3 and +6 position DNA-contact residues in the C-terminal finger of ADRl do not appear to be critical for DNA binding (see Figure 5 ) . The well-defined nature of the side chains at the +3 position in both ADRl finger structures makes this residue in the ADR1-DNA models particularly interesting. Superposition models highlighting this position in both ADRl fingers are shown in Figure 6. Superpositioning places His118, the +3 residue in ADRlb, proximal to the middle base of the subsite (A9). Without any adjustments to the side-chain conformation of the solution structure, the imidazolering is in excellentH-bonding distance and geometry with the N7 atom of the adenine (adenine nitrogen-toimidazole proton distance of 1.7 A). An almost identical interaction was previously observed for the histidine in the +3 position of Zif268 finger 2 and the N7 of a guanine. The +3 position in ADRla is Leu146, and the middle base in its DNA subsite is a thymidine (T6). The superposition model for PAPA places Leu 146 in proximity to T6, as shown in Figure 6. However, if this finger is in the same relative orientation to its site as the Zif268 fingers, a steric clash is indicated between one of the methyl groups of Leu146 and
the thymidine methyl group, with leucine methyl proton to thymidine methyl proton distances of 1.1 A (compare to the van der Waals radius of a proton of 0.9 A). The clash could not be relieved by altering the x2 for the leucine side chain. Therefore, superpositioning suggests that the ADR 1a finger must be adjusted within the major groove at this position, relative to Zif268. Possible results of such an adjustment would be favorable van der Waals interactions and/or hydrophobicinteractionsbetween the methyl groups of Leu 146 and the methyl group of T6. Substitution of a purine at this position in the PAPA-DNA model would remove these favorableinteractions,consistent with mutagenesis results that showed that a cytosine is accommodated at this position, while substitution with either purine base led to a loss of DNA binding (Thukral et al., 1992). The lack of effect of substituting the +6 position Arg149 residue with alanine in the ADRla finger is perhaps the most curious result of the alanine mutagenesis study. The ADRl b finger also contains an arginine at this position which changeof-specificity mutation indicatesis contacting guanine (Thukral et al., 1992). Substitution of Argl21 with alanine was one of the most detrimental alanine mutations made, suggesting that this interaction contributes significantly to the overall binding free energy. On the other hand, the equivalentADRla residue, Arg149, is not crucial for DNA affinity (Thukral et al., 1991b). Moreover, unlike the subsites for Zif268 fingers 1 and 3 and ADRlb, the ADRla binding site contains a thymidine at the site where R149 is expected to interact, rather than a guanine. Wild-type ADRl will accommodateany other base at this position, strongly suggesting that Arg149 does not directly contact the base (Thukral et al., 1992). Since there is no detectable difference in the structures of ADRl b and PAPA at the +6 positions, we looked for a distal effect. As described above, the superposition model of PAPA-DNA indicates that the ADRl a finger must have its a-helix adjusted in the major groove of the DNA due to the bulky side chain at its +3 position. We suggest that this, in turn, affects the ability of the +6 position Arg149 to make a base contact. Such a neighboring effect of one putative DNA-contact residue on another is consistent with mutagenesis experiments on another zinc finger protein, SP1 (Desjarlais & Berg, 1992). This hypothesis is testable experimentally: substitution of a smaller residue for Leu146 should result in an increased interaction for Arg 149. Recent mutagenesis experiments in
Structure of a His-Xd-His Zinc Finger Domain which Leu146 was substituted with an alanine side chain are indeed consistent with this hypothesis. Unlike wild-type ADR1, the mutant protein has a preference for a guanine at position 5 , the position that Arg149 is predicted to contact (C. Cheng and E. T. Young, personal communication).
CONCLUSIONS Our goal has been to understand the functional aspects of ADRl DNA binding in terms of its structure. Toward this end, we have determined and refined the solution structures of zinc finger domains from ADRl individually. The relevance of this approach is borne out in the finding that the refined structures obtained in the absence of DNA are remarkably similar to the structures of zinc fingers bound to DNA determined by X-ray diffraction. Thus, the solution structures are excellent models from which to derive structure-function information about the DNA-binding domain of ADRl in particular and of (Cys2, Hisz) zinc finger proteins in general. The results for the PAPA mutant indicate that the ADRla sequence is capable of existing in a canonical zinc finger structure that is very similar to those described for ADRl b and other (Cys2, His2) fingers. However, the observation that Ala133 in PAPA exists in a conformation that is disallowed for proline indicates that the wild-type domain must have an altered conformation in its N-terminal region. This is consistent with our original report on wild-type ADRla in which it was stated that the NOES expected for a classical P-sheet were not observed (Pgrraga et al., 1988). Of the two prolines in ADRl a, Prol 3 1 has been implicated as critical for DNA binding since its replacement with alanine resulted in a 100-fold decrease in DNA-binding affinity in a functional two-finger ADRl construct (Thukral et al., 1991b). Proline is found with high frequency at this position in multifinger proteins, suggesting that this result may be generalizable. Solution NMR studies of two different multiple zinc finger domains, neither of which contains proline in its linkers, have revealed that the dispositionsof individual fingers in these constructs are not well-defined in the absence of DNA (Omichinskietal., 1992;Nakasekoetal., 1992). The presence of proline in a linker may serve to limit the range of orientations of two fingers with respect to each other in the free protein. There are prolines in fingers 2 and 3 of the Zif268, both of which are in the trans conformation, with (4, #) values of (--60, -10) in the protein-DNAcomplex. Thisconformation is one of the two low-energy conformations for prolines (Cantor & Schimmel, 1980), so it is likely that the prolines are in a similar conformation in the free protein. Although the conformation observed for the two Zif268 prolines is accessible to an alanine residue, an alanine is much less restricted in its allowable backbone conformations. Thus, the substitution of Prol 3 1with alanine may increase the conformational entropy of the free protein, and the 3 kcal/mol loss of binding free energy measured for the P131A ADRl mutant may reflect an increased loss of entropy when the contiguous zinc fingers bind to DNA. A second example of loss of DNA affinity on replacing an analogous proline in the linker of fingers 1 and 2 of TFIIIA has recently been reported (Choo & Klug, 1993). Several other conclusions that can be drawn from the PAPA solution structure serve to explain earlier observations. First, the residue that follows the second cystine ligand has a positive 4 angle, similar to the refined ADRlb structure (Hoffman et a[., 1993a). This may explain the high degree of conservation of glycine at this position and has implications to the design of novel zinc finger domains, as a glycine at this position should better favor the correct folded structure. Second, the
-
Biochemistry, Vol. 33, No. 15, 1994 4469 extra residue between the histidine ligands in ADRla has only a local effect on the conformation of the domain, consistent with the reported structure of another His-X4-His finger domain (Kochoyan et al., 1991). Comparison of the structure of PAPA and ADRl b in this region shows that the overriding force is the geometry of the zinc ligands, which is nearly the same in the two structures, regardless of the different number of residues between them. Thus, it is the backbone conformation, not the metal-ligand conformation, that adjusts in response to different loop lengths. Third, a side-chain-toside-chain H-bond is strongly indicated between two surfaceaccessible, hydrophilic residues in PAPA, namely, Arg143 and Asp145. An analogous H-bond is observed in the DNAbound form of Zif268, but it was somewhat unexpected to find such a H-bond existing in a domain free in solution. The fact that a similar H-bond is not indicated by the data for ADRlb which contains an analogous arginine but a glutamic acid in the place of Asp145 suggests that there is a delicate balance between the gain in enthalpy and the loss of entropy for this H-bond such that it does not form spontaneously when the longer acidic side-chain is present. The model created for the interaction between ADRl finger domains and their respective DNA subsites simply by superpositioning the refined peptide solution structures onto the Zif268-DNA cocrystal structure predicts many of the known DNA-binding properties of the ADRl fingers and may therefore serve to reveal new insights as well. In particular, the model offers an explanation for the functional properties of the +3 and +6 positions in ADRla, which mutagenesis experiments had indicated were different from their ADRl b counterparts. Specifically,the model suggests that the ADRla finger-DNA geometry must be somewhat different from that seen for the three Zif268 fingers, since the analogous geometry results in a steric clash between methyl groups of the +3 Leu146 residue with the methyl group of T6. A slight adjustment of the finger relative to the major groove will alleviate steric overlap and could allow for a favorable interaction between the two methyl groups. This predicted adjustment of ADRla relative to the DNA may also offer an explanation for the observation that the arginine at the +6 position (Arg149) does not appear to have any interactions with DNA. Taken together, the model explains the mutagenesis results and points to what is certain to become a recurring theme of zinc finger-DNA interaction, namely, the non-independence of neighboring DNA-contacting residues (Desjarlais & Berg, 1992; Bernstein et al., 1993). In conclusion, the refined structures of the two zinc finger domains from ADRl have afforded insights into the structural features of this class of DNA-binding motifs. In conjunction with the extensive information available from mutagenesis and functional studies of ADRl, these structural models also offer insights into the structure-function relationships that are central to the sequence-specific DNA recognition between ADRl and the UASl site. The hypotheses offered here on the basisofthe ADR1-UAS1 modelarecurrently being tested experimentally, both by further mutagenesis studies and by a direct structural determination of the complex formed between the DNA-binding domain of ADRl and the UASl DNA sequence. This combined approach will undoubtedly yield further insights in the future.
ACKNOWLEDGMENT The authors thank Mia Schmiedeskamp and Ted Young for helpful discussions and critical reading of the manuscript, David English for sharing his results on the C140A peptide,
4470 Biochemistry, Vol. 33, No. 15, 1994 Cheng Cheng for sharing his results on Leu146 mutants, and Nikola Pavletich and Carl Pabo for the Zif268-DNA crystal structure coordinates.
SUPPLEMENTARY MATERIAL AVAILABLE One figure showingthe region of the HzO NOESY spectrum of PAPA and two tables giving distance constraints used to calculate the solution structure of PAPA using DIANA and angle constraints used to generate the structure (9 pages). Ordering information is given on any current masthead page.
REFERENCES Adman, E. T., Sieker, L. C., & Jensen, L. H. (1991) J.Mol. Biol. 217, 337-352. Bernstein, B. E., Hoffman, R. C., Horvath, S. J., Herriott, J. R., & Klevit, R. E. (1993) Ann. N.Y. Acad. Sci. (in press). Brown, R. S., Sander, C., & Argos, P. (1985) FEBS Lett. 186, 27 1-274. Cantor, C. R., & Schimmel,P. S. (1980) Biophysical Chemistry, Part 1, W. H. Freeman and Co., San Francisco. Choo, Y., & Klug, A. (1993) Nucleic Acids Res. 21,3341-3346. Denis, C. L., & Young, E. T. (1983) Mol. Cell. Biol. 3,360-370. Denk, W., Wagner, G., Rance, M., & Wiithrich, K. (1985) J. Magn. Reson. 62, 350-355. Desjarlais, J. R., & Berg, J. M. (1992) Proc. Natl. Acad. Sci. U.S.A.89, 7345-7349. El-Baradi, T., & Peiler, T. (1991) Mech. Den 35, 155-169. Hoffman, R. C., Horvath, S., & Klevit, R. E. (1993a) Protein Sci. 2, 951-965. Hoffman, R. C., Xu, R., Klevit, R. E., & Herriott, J. R. (1993b) J . Magn. Reson., Ser. B 102, 61-72. Klevit, R. E., Herriott, J. R., & Horvath, S. J. (1990) Proteins 7, 215-226. Kochoyan, M., Havel, T. F., Nguyen, D. T., Dahl, C. E., Keutmann, H. T., & Weiss, M. A. (1991) Biochemistry 30, 337 1-3 386. Kurtz, T. E. (1963) Basic Statistics, Prentice-Hall, Inc., Englewood CLiffs, NJ.
Bernstein et al. Lee, M. S., Cavanagh, J., & Wright, P. E. (1989a) FEBS Lett. 254, 159-164. Lee, M. S., Gippert, G. P., Soman, K. V., Case, D. A., & Wright, P. E. (1989b) Science 245, 635-637. Lee, M. S., Mortishire-Smith, R. J., & Wright, P. E. (1992) FEBS Lett. 309, 29-32. Michael, S. F., Kilfoil, V. J., Schmidt, M. H., Amann, B. T., & Berg, J. M. (1992) Proc. Natl. Acad. Sci. U.S.A. 89, 47964800. Miller, J., McLachlan, A. D., & Klug, A. (1985) EMBO J. 4, 1609-1 614. Mueller, L. (1987) J. Magn. Reson. 72, 191-197. Nakaseko, Y., Neuhaus, D., Klug, A., & Rhodes, D. (1992) J. Mol. Biol. 228, 619-636. Omichinski, J. G., Clore, G. M., Appella, E., Sakaguchi, K., & Gronenborn, A. M. (1990) Biochemistry 29, 9324-9334. Omichinski, J. G., Clore, G. M., Robien, M., Sakaguchi, K., Appella, E., & Gronenborn, A. M. (1992) Biochemistry 31, 3907-3917. PArraga, G. E., Horvath, S. J., Eisen, A., Taylor, W. E., Hood, L., Young, E. T., & Klevit, R. E. (1988) Science 241, 14891492. PArraga, G. E., Horvath, S. J., Hood, L., Young, E. T., & Klevit, R. E. (1990) Proc. Narl. Acad. Sci. U.S.A.87, 137-141. Pavletich, N. P., & Pabo, C. 0. (1991) Science 252, 809-817. Richardson, J. S. (198 1) The Anatomy and Taxonomy of Protein Structure, in Advances in Protein Chemistry (Anfinsen, C. B., Edsall, J. T., & Richards, F. M., Eds.) Vol. 34, pp 167-339, Academic Press, Inc., New York, NY. Thukral, S. K., Eisen, A,, & Young, E. T. (1991a) Mol. Cell. Biol. 11, 1566-1577. Thukral, S. K., Morrison, M. L., & Young, E. T. (199 1b) Proc. Natl. Acad. Sci. U.S.A. 88, 9188-9192. Thukral, S. K., Morrison, M. L., & Young, E. T. (1992) Mol. Cell. Biol. 12, 2784-2792. Wuthrich, K. (1986) NMR of Proteins and Nucleic Acids, John Wiley & Sons, Inc., New York, NY. Xu, R. X., Horvath, S. J., & Klevit, R. E. (1991) Biochemistry 30, 3365-3371.
